How to log a error to Stackdriver Error Reporting via Stackdriver Logging in Python

Question

I've been using Stackdriver Logging for a long time and now I'd like to also take advantage of Error Reporting. I'd prefer to use Python's logging mechanism and scrape exceptions out of a log file if possible (for various reasons) rather than using the error_reporting library. That being said, the documentation is very confusing. For example the documentation says: https://cloud.google.com/error-reporting/docs/setup/compute-engine#log_exceptions

First, install the fluent-logger-python library:

sudo pip install google-cloud-error-reporting --upgrade

which leads me to believe that google-cloud-error-reporting is a fork of or related to fluent-logger-python however, when I initialize google-cloud-error-reporting it directly calls out to the GCE metadata server rather than connecting to the local fluentd. Are these two unrelated packages or is the documentation wrong or misleading? If I send JSON formatted exceptions to fluentd or to a log file monitored by fluentd will error reporting understand them?

Thanks for any clarifications

Answer 1

The documentation is wrong.
TL;DR You have to output something that looks like https://cloud.google.com/error-reporting/docs/formatting-error-messages

Here's my solution:

#Parse raw log entries to expose severity field so that
#StackDriver log viewer can properly categorize (and so we can filter)
<source>
  @type tail
  path /var/log/conductor
  pos_file /var/log/td-agent/conductor.pos
  format multiline
  format_firstline /\d{4}-\d{2}-\d{2} \d{2}:\d{2}:\d{2}/
  format1 /^(?<message>(?<time>\d{4}-\d{2}-\d{2} \d{2}:\d{2}:\d{2}),\d*\s*[a-zA-Z_]*\s*(?<severity>[A-Z]*).*)/
  read_from_head true
  multiline_flush_interval 3s
  tag conductor.app
</source>

#Add hostname field
<filter conductor.app>
  @type record_transformer
  <record>
    hostname ${hostname}
  </record>
</filter>

#Filter and tag log entries of severity ERROR or CRITICAL
<match conductor.app>
  @type rewrite_tag_filter
  rewriterule1 severity ERROR|CRITICAL conductor.err
  rewriterule2 severity .+ conductor.info
</match>

#Process entries with tracebacks differently than those without
<match conductor.err>
  @type rewrite_tag_filter
  rewriterule1 message .*Traceback conductor.err.traceback
  rewriterule2 message .+ conductor.err.message
</match>

#Parse out the traceback
<match conductor.err.traceback>
  @type parser
  key_name message
  format multiline
  format1 /^(?<message>(?<time>\d{4}-\d{2}-\d{2} \d{2}:\d{2}:\d{2}),\d*\s*(?<log>[a-zA-Z_]*)\s*(?<severity>[A-Z]*).*(?
<traceback>Traceback .*))/
  tag conductor.err.traceback.report
</match>

#Format traceback reports
<filter conductor.err.traceback.report>
  @type record_transformer
  <record>
      serviceContext {
        "service": "${record[\"log\"]}"
      }
      message ${record["traceback"]}
  </record>
  remove_keys traceback
</filter>

#Process errors that don't have tracebacks
<match conductor.err.message>
  @type parser
  key_name message
  format multiline
  format1 /^(?<message>(?<time>\d{4}-\d{2}-\d{2} \d{2}:\d{2}:\d{2}),\d*\s*(?<log>[a-zA-Z_]*)\s*(?<severity>[A-Z]*):\s*(?<report>.*))/
  tag conductor.err.message.report
</match>

#For errors without tracebacks we have to stub out some fields that
#error reporting requires, but we don't have
<filter conductor.err.message.report>
  @type record_transformer
  <record>
      serviceContext {
        "service": "${record[\"log\"]}"
      }
      message ${record["report"]}
      reportLocation {
        "filePath": "None",
        "lineNumber": 0,
        "functionName": "None"
      }
  </record>
</filter>

#Send to StackDriver logging!
<match conductor.**>
  @type google_cloud
  buffer_chunk_limit 2M
  flush_interval 5s
  max_retry_wait 300
  disable_retry_limit
  num_threads 8
</match>